⚡️🦾 Compute Scaled. Bandwidth Fell Behind. Now What?

A Newsletter for Entrepreneurs, Investors, and Computing Geeks

Author

Emily Eisenblätter
July 28, 2025

Happy Monday! This week’s deep dive highlights the bandwidth bottleneck in modern computing and introduces a material that could help overcome it.

In our spotlights, we highlight a foundational discovery in quantum physics that introduces a reversible framework for managing entanglement and trace Perplexity’s rise.

We also cover major headlines across AI, semiconductors, quantum, photonics, data centers, and cloud, along with curated readings on chip packaging, Apple’s AI training stack, and sustainable cooling. Funding news was unusually quiet, but we’ve included five notable later-stage rounds.

And in our bonus section, we break down last week’s news on OpenAI and SoftBank’s $500B Stargate project, which appears to be facing setbacks due to internal friction.

Deep Dive: Compute Scaled. Bandwidth Fell Behind. Now What?

Over the past 20 years, improvements in compute speed have outpaced advances in bandwidth by a factor of roughly 1,000. The performance of processors has grown rapidly, but the rate at which data can be transferred between them has increased much more slowly. As a result, communication across systems has become a central limitation in modern computing.

Why Bandwidth Is a Bottleneck

Processing units can only operate efficiently if data moves between them without delay. Bandwidth defines how much information can be exchanged per second. When it falls short, even powerful processors are left waiting. This is particularly relevant for large-scale workloads like AI, where vast amounts of data must be exchanged in real time.

Optical Interconnects

To move data quickly and efficiently, most systems use optical interconnects that transmit signals with light instead of electricity. Photons enable lower energy loss and higher speed, which is why optical fiber underpins today’s internet and why the same approach is already common in data centers and expanding further into compute infrastructure.

The Role of Photonic Chips

Photonic integrated chips (PICs) serve as the bridge between the electronic and optical domains. They translate signals from electrons to photons and back. Most PICs today are made from silicon. While silicon is widely used and scalable, it no longer meets the performance requirements needed to overcome the growing bandwidth bottleneck in modern systems.

A Potential Solution

Thin-Film Lithium Niobate (TFLN) is a material with strong optical properties. It enables optical data transmission up to four times faster and four times more energy-efficient than conventional electrical links. This makes it a promising option for future communication systems where both high bandwidth and energy efficiency are critical.

If you are interested in how the manufacturing of this material can be scaled for industrial use, read our interview with Lightium co-founder and CRO Frédéric Loizeau.

Other companies working with this material are Q.ANT, HyperLight, and CCRAFT.

Spotlights

Entangled quantum states can be reversibly transformed using an “entanglement battery” that stores and releases entanglement like a power supply. This simple but powerful idea could make quantum technologies more efficient and extends to quantum thermodynamics, hinting at a deeper set of rules for managing quantum resources.

Perplexity’s $18B rise reflects a smart bet: build on top of existing LLMs rather than compete with them. The article traces how Aravind Srinivas turned a prototype into a widely used answer engine, why Google couldn’t easily copy it, and how Perplexity is now expanding through telecom partnerships and its AI-native browser to challenge Google’s dominance in how information is accessed online.

Headlines

Last week’s headlines cover major moves in AI infrastructure, new materials and packaging in semiconductors, a burst of activity in quantum computing, and a quieter week for neuromorphic tech.

 🤖 AI

🦾 Semiconductors

⚛️ Quantum Computing

🧠 Neuromorphic Computing

⚡️ Photonic / Optical Computing

☁️ Cloud

Readings

This week’s reading list covers Apple’s AI training methods, advances in chip packaging and quantum error correction, and new cooling strategies for data centers in the AI era.

 🤖 AI

🦾 Semiconductors

The Rise of Panel-Level Packaging (Semiconductor Engineering) (11 mins)

VLSI 2025 Roundup (SemiAnalysis) (14 mins)

⚛️ Quantum Computing

💥 Data Centers

☁️ Cloud

Funding News

Last week’s funding news was unusually quiet, with no early-stage activity. Let us know if we missed anything.

Amount

Name

Round

Category

$40M

xLight

Series B

Semiconductors

$59M

Lumotive

Series B

Photonic / Optical

“several ¥100M”

SpinQ

Series B

Quantum

$131M

Armada

Venture Round

Data Centers

$835M

5C

Venture Round (incl. Debt)

Data Centers

Bonus: What’s Happening With the $500B AI Megaproject Stargate?

What Is Stargate?

Stargate is an AI infrastructure initiative announced in early 2025 by OpenAI, SoftBank, Oracle, and other partners. The project was positioned as a $500 billion effort to massively expand compute capacity, with plans for multiple large-scale data centers. The overall goal was to build over 5 gigawatts of capacity to support future AI models.

What Is The Status Quo?

Last week, reports revealed that Stargate is already running into delays and internal disagreements. Instead of breaking ground on multiple hyperscale campuses, the near-term focus has shifted to a smaller data center in Ohio, targeted for completion by the end of 2025. The slowdown is reportedly due to friction between OpenAI and SoftBank over site selection and operational control.

Last Week’s Articles

❤️ Love these insights? Forward this newsletter to a friend or two. They can subscribe here.